bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

1 GOT1 Inhibition Primes Pancreatic Cancer for Ferroptosis through the 2 Autophagic Release of Labile Iron 3 4 Daniel M. Kremer1,2, Barbara S. Nelson3, Lin Lin1, Emily L.Yarosz4, Christopher J. 5 Halbrook1, Samuel A. Kerk3, Peter Sajjakulnukit3, Amy Myers1, Galloway Thurston1, 6 Sean W. Hou1, Eileen S. Carpenter5, Anthony C. Andren1, Zeribe C. Nwosu1, Nicholas 7 Cusmano1, Stephanie Wisner1, Johanna Ramos1, Tina Gao1, Stephen A. Sastra6,7, 8 Carmine F. Palermo6,7, Michael A. Badgley6,7,8, Li Zhang1, John M. Asara9,10, Marina 9 Pasca di Magliano11,12,13, Yatrik M. Shah1,5,11, Howard C. Crawford1,11, Kenneth P. 10 Olive6,7, Costas A. Lyssiotis1,5,11‡ 11 12 1 Department of Molecular and Integrative Physiology, University of Michigan, Ann Arbor, MI 48109, USA 13 2 Graduate Program in Chemical Biology, University of Michigan, Ann Arbor, MI 48109, USA 14 3 Graduate Program in Cancer Biology, University of Michigan, Ann Arbor, MI 48109, USA 15 4 Immunology Graduate Program, University of Michigan, Ann Arbor, MI 48109, USA 16 5 Department of Internal Medicine, Division of Gastroenterology and Hepatology, University of Michigan, Ann Arbor, 17 MI 48109, USA 18 6 Division of Digestive and Liver Diseases, Department of Medicine, Columbia University Medical Center, New York, 19 NY 10032, USA 20 7 Herbert Irving Comprehensive Cancer Center, Columbia University Medical Center, New York, NY 10032, USA 21 8 Department of Pathology, Columbia University Medical Center, New York, NY 10032, USA 22 9 Division of Signal Transduction and Mass Spectrometry Core, Beth Israel Deaconess Medical Center, Boston, MA, 23 USA 02115 24 10 Department of Medicine, Harvard Medical School, Boston, MA, USA 02115 25 11 Rogel Cancer Center, University of Michigan, Ann Arbor, MI 48109, USA 26 12 Department of Surgery, University of Michigan, Ann Arbor, MI 48109, USA 27 13 Department of Cell and Developmental Biology, University of Michigan, Ann Arbor, Michigan 48109, USA. 28 ‡ Correspondence: [email protected] 29 30 Highlights 31  PDA exhibit varying dependence on GOT1 for in vitro and in vivo growth. 32  Exogenous cystine, glutathione synthesis, and lipid antioxidant fidelity are 33 essential under GOT1 suppression. 34  GOT1 inhibition sensitizes pancreatic cancer cell lines to ferroptosis. 35  GOT1 inhibition represses anabolic metabolism and promotes the release of iron 36 through autophagy. 37 38 Summary 39 Pancreatic ductal adenocarcinoma (PDA) is one of the deadliest solid malignancies, 40 with a 5-year survival rate at ten percent. PDA have unique metabolic adaptations in 41 response to cell-intrinsic and environmental stressors, and identifying new strategies to 42 target these adaptions is an area of active research. We previously described a 43 dependency on a cytosolic aspartate aminotransaminase (GOT1)-dependent pathway 44 for NADPH generation. Here, we sought to identify metabolic dependencies induced by 45 GOT1 inhibition that could be exploited to selectively kill PDA. Using pharmacological 46 methods, we identified cysteine, glutathione, and lipid antioxidant function as metabolic 47 vulnerabilities following GOT1 withdrawal. Targeting any of these pathways was 48 synthetic lethal in GOT1 knockdown cells and triggered ferroptosis, an oxidative, non- 49 apoptotic, iron-dependent form of cell death. Mechanistically, GOT1 inhibition promoted 50 the activation of autophagy in response to metabolic stress. This enhanced the 51 availability of labile iron through ferritinophagy, the autolysosome-mediated degradation

1 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

52 of ferritin. In sum, our study identifies a novel biochemical connection between GOT1, 53 iron regulation, and ferroptosis, and suggests the rewired malate-aspartate shuttle plays 54 a role in protecting PDA from severe oxidative challenge. 55 56 Keywords: Pancreatic ductal adenocarcinoma, GOT1, cysteine, glutathione, GPX4, 57 NADPH, redox, reactive oxygen species, ferroptosis, labile iron, autophagy, 58 ferritinophagy, metabolic dependencies 59 60 Introduction 61 Pancreatic ductal adenocarcinoma (PDA) is a notoriously lethal disease. This stems 62 from late-stage diagnosis and the lack of effective therapies1. A defining feature of PDA 63 is the extensive fibroinflammatory reaction that not only regulates cancer initiation, 64 progression, and maintenance, but also promotes its therapeutic resilience2,3,4. Vascular 65 collapse, impaired perfusion, and hypoxia accompany this desmoplastic reaction to 66 produce a nutrient-deprived and harsh tumor microenvironment3,5. PDA cells reprogram 67 their nutrient acquisition and metabolism to support survival and growth under these 68 conditions6,7. 69 70 Our previous work demonstrated that PDA rewire the malate-aspartate shuttle to 71 generate reduced nicotinamide adenine dinucleotide phosphate (NADPH), a major 72 currency for biosynthesis and redox balance (Figure 1A)8. The malate-aspartate shuttle 73 canonically functions to transfer reducing equivalents in the form of NADH from the 74 cytosol into the mitochondria to facilitate oxidative phosphorylation (OxPHOS). In PDA, 75 we found the mitochondrial aspartate aminotransaminase (GOT2) is the primary 76 anaplerotic source for alpha-ketoglutarate (αKG) and generates aspartate. Aspartate is 77 then transferred to the cytosol and transaminated to produce oxaloacetate (OAA) by the 78 cytosolic aspartate aminotransaminase (GOT1). OAA is reduced to malate by cytosolic 79 Malate Dehydrogenase (MDH1) and is then oxidized by Malic Enzyme 1 (ME1) to 80 generate NADPH, which is utilized to support redox balance and proliferation in PDA8. 81 Furthermore, we demonstrated that this non-canonical pathway was orchestrated by 82 mutant KRAS, the signature oncogenic driver of PDA. Specifically, mutant Kras led to 83 the transcriptional upregulation of GOT1 and the concurrent repression of GLUD1, 84 which drives an alternative metabolic pathway. Indeed, a higher expression of GOT1 to 85 GLUD1 is associated with worse patient outcome9. Thus, in an effort to target this 86 rewired metabolic pathway, we have placed our focus on GOT1. And, notably, we and 87 others recently identified drug scaffolds that may serve as leads in the development of a 88 clinical grade GOT1 inhibitor 10,11,12,13. 89 90 Herein, we present a detailed analysis of GOT1 dependence across a large panel of 91 PDA cell lines and specimens. We demonstrate that GOT1 sensitivity varies among the 92 cultures in this panel, is dispensable in non-transformed human lines, and that GOT1 93 inhibition stunted growth in tumor xenograft models. In GOT1 dependent contexts, 94 GOT1 inhibition blocked progression through the cell cycle, leading to cytostasis. Thus, 95 we then sought to characterize metabolic dependencies following GOT1 withdrawal that 96 could be exploited to selectively kill PDA14,15. Examination of a targeted library of 97 metabolic inhibitors in GOT1 knockdown cells led to the discovery that that exogenous

2 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

98 cystine was essential for viability following chronic GOT1 suppression. Cystine is used 99 for reduced glutathione (GSH) biosynthesis, which mediates protection against lipid 100 oxidation. GOT1 knockdown in combination with inhibitors of glutathione synthesis or 101 lipid antioxidant machinery led to cell death. We characterized this as ferroptosis: an 102 oxidative, non-apoptotic, and iron-dependent form of cell death. We then determined 103 that GOT1 withdrawal promoted a catabolic cell state that resulted in decreased 104 OxPHOS and the activation of autophagy and ferritinophagy16. Ferritinophagy primed 105 GOT1 knockdown cells for ferroptosis by increasing labile iron pools. Together with the 106 modulation of NADPH and GSH levels, our study demonstrates that GOT1 inhibition 107 promotes ferroptosis sensitivity by promoting labile iron and illustrates how the rewired 108 malate-aspartate shuttle participates in the maintenance of both antioxidant and 109 energetic balance. 110 111 Results 112 GOT1 dependent PDAs require GOT1 for cell cycle progression and proliferation 113 To examine GOT1 dependence in a large panel of PDA cell lines and primary 114 specimens with temporal control, we developed doxycycline (dox)-inducible short 115 hairpin (sh)RNA reagents (iDox-sh) that target the coding and 3’UTR regions of GOT1 116 (sh1 and sh3), or scramble (shNT). shRNA activity was examined phenotypically by 117 assessing colony formation and protein levels following dox treatment (Figures 1B,C 118 and S1A,B). We also measured aspartate levels as a biochemical readout for GOT1 119 inhibition (Figure S1C). We then used these iDox-shRNA constructs to assess GOT1 120 sensitivity across a large panel of PDA lines and primary specimens (indicated with the 121 UM# designation)17 (Figure 1D and S1D). GOT1 knockdown significantly impaired 122 colony formation in 12 of 18 cell lines in the panel (Figures 1D and S1A,B). 123 Indeed, in PL45, Capan-1, and Pa-Tu-8902 colony formation was severely diminished 124 following GOT1 knockdown with both hairpins, and this was independent of dox-effects 125 (Figure S1D). Further, the response to GOT1 knockdown did not correlate with 126 common mutations associated with PDA (Figure S1E) or expression of malate- 127 aspartate shuttle enzymes (Figure S1F). To test the specificity of GOT1 against PDA, 128 we extended our cell panel to non-transformed human lines. We found human 129 pancreatic stellate cells (hPSC), human lung fibroblasts (IMR-90), and human non- 130 transformed pancreatic exocrine cells (hPNE) were minimally affected upon GOT1 131 knockdown, in agreement with previous results, suggesting that this pathway may be 132 dispensable in non-transformed cells (Figures 1E and S1G)8,12. Together, these data 133 demonstrate many PDA cell lines require GOT1 for growth while non-transformed cell 134 lines do not, highlighting a potential therapeutic window. 135 136 To determine the relevance of GOT1 in vivo, we examined the effect of GOT1 inhibition 137 on established PDA tumors. PDA cells were implanted subcutaneously into the flanks or 138 orthotopically into the pancreas of immunocompromised mice and allowed to establish 139 for 7 days prior to GOT1 inhibition. GOT1 sensitive cell lines exhibited profound growth 140 inhibition upon induction of GOT1 knockdown with dox (Figures 1F,G), results that 141 were consistent with previous studies that employed constitutive shGOT18. Parallel 142 studies with shNT tumors indicated that the effect was independent of dox exposure

3 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

143

4 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

Figure 1. GOT1 dependence is a feature of some PDA cell lines, matched by G1 cell cycle arrest and inhibition of proliferation. A) Model of the rewired malate-aspartate shuttle in PDA. B-C) Representative colony formation assay (B) and immunoblot analysis (C) of Pa-Tu-8902 cells stably expressing iDox-shRNA constructs following 10 days GOT1 knockdown. shRNAs target the coding region of GOT1 (sh1), or the 3’UTR region of GOT1 (sh3). Parental (parent) and scramble (shNT) conditions are also displayed (n=3). Vinculin (VCL) was used as a loading control. D) Relative colony number across a panel of PDA cell lines in response to GOT1 knockdown by sh1 (black/blue) or sh 3 (grey/red). Assays were run 10-15 days (n=3), n.s. denotes non-significant. E) Relative cell number of immortalized, non-transformed, cell lines after 1, 3, or 5 days, normalized to day 1 (n=3). F) Growth of subcutaneous xenograft tumors from 3 PDA cell lines stably expressing iDox-shGOT1 or shNT. Dox or mock conditions were administered 7 days following implantation. Treatment with dox (red) or vehicle (black) (BxPC-3 n= 8, MIA PaCa-2 n=6, Pa-Tu-8902 n=6 per arm). G) Orthotopic xenograft tumor growth from Pa-Tu-8902 iDox-shGOT1 stable cell lines co- expressing firefly luciferase (FLuc) n=5 and n=6 mice were used for vehicle and dox cohorts respectively. H-I) Immunoblots for GOT1 (H) and relative aspartate levels (I) measured by liquid-chromatography tandem mass spectrometry, normalized to -dox (n=5). J) Histology of BxPC-3 iDox-shGOT1 subcutaneous xenograft tumors from vehicle- or dox-treated mice. H&E, Hematoxylin and Eosin, CC3, cleaved caspase 3. Scale bars represent 50µm. K) Cell cycle distribution of Pa-Tu-8902 iDox-GOT1 sh1 upon 1,3, or 5 days of dox treatment. Significance values are in relation to iDox-shGOT1 mock (n=3). L) Proliferation kinetics following GOT1 knockdown. Cells were untreated (black), dox was added to untreated cells (blue), pre- treated with dox and chronically exposed to dox (grey), or released from dox pretreated cells (red). Relative cell number at day 5 normalized to day 1 is displayed, (n=3). Error bars represent mean ± SD. Two-tailed unpaired T-test or 1-way ANOVA: Non-significant P > 0.05 (n.s. or # as noted), P ≤ 0.05 (*), ≤ 0.01 (**), ≤ 0.001 (***), ≤ 0.0001 (****). See also Figure S1. 144 (Figure S1H). GOT1 knockdown was demonstrated by immunoblot analysis on 145 homogenized tumor tissue (Figures 1H and S1I) and biochemically via the induction of 146 aspartate (Figure 1I). Contrasting our in vitro studies in which GOT1 was knocked down 147 for five days (Figure S1C), the changes in aspartate abundance described in this 148 experiment reflect the whole tumor metabolome after thirty days of dox treatment. While 149 immunostaining for GOT1 indicate potent knockdown in the PDA cell compartment 150 (Figures 1J), it is unknown how long-term suppression of GOT1 would influence non- 151 cell autonomous metabolism in this xenograft study. Tumor growth suppression was 152 confirmed at the molecular level by a decrease in Ki-67, a marker for proliferation. 153 GOT1 knockdown tumors exhibited minimal staining for cleaved caspase 3 (CC3), a 154 marker for apoptosis. Thus, these proliferative defects were independent of apoptosis, 155 indicating GOT1 inhibits tumor proliferation, rather than, inducing cell death (Figure 1J). 156 157 To test the hypothesis that GOT1 inhibition is cytostatic, we examined the effect of 158 GOT1 knockdown on cell cycle progression. Knockdown led to a higher distribution of 159 cells in G1 phase versus the S and G2 phases, indicating that the majority of cells are in 160 G1 cell cycle arrest following GOT1 inhibition (Figures 1K and S1J). Moreover, the 161 effect of GOT1 knockdown was reversible, as cells regained proliferative capacity upon 162 removal of genetic inhibition (Figures 1L and S1K). Overall, PDA display a spectrum of 163 sensitivity to GOT1 where GOT1 inhibition is cytostatic. 164 165 Limiting exogenous cystine potentiates GOT1 inhibition and elicits a cytotoxic 166 effect 167 Because GOT1 inhibition is cytostatic, we sought to identify metabolic dependencies 168 induced by knockdown that could be targeted to selectively kill PDA14,15. Our previous

5 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

169 work indicated that inhibiting glutamine metabolism enhances sensitivity to reactive 170 oxygen species (ROS)8,9,18. Thus, we assessed the sensitivity of PDA cells to a panel of 171 chemotherapeutic agents and inhibitors of antioxidant pathways after five days of GOT1 172 knockdown and three days of drug treatment (Figures 2A,B). 173 174 GOT1 inhibition was protective when combined with some inhibitors demonstrated by 175 increased area under the curve values (AUC), a measure of drug sensitivity (Figures 176 2B and S2A). Three of the five top desensitizing agents were chemotherapies, in 177 agreement with previous observations18. Since these chemotherapies work through 178 disrupting DNA replication of rapidly dividing cells, we speculate that the decreased 179 sensitivity may occur due to the GOT1 growth-suppressive phenotype (Figures 1J,K). 180 By contrast, GOT1 knockdown sensitized PDA cells to erastin (Figure 2B), and the 181 effect was stronger after 24 hours of treatment (Figure 2C). Erastin is an inhibitor of 19,20 - 182 xCT , a component of the system xc cystine/glutamate antiporter which transports 183 cystine into cells in exchange for glutamate. Cystine, the oxidized dimer of cysteine, is 184 reduced to cysteine upon entering the cell where it can contribute to the synthesis of 185 GSH and proteins, among numerous other biochemical fates. 186 187 We then tested the combination of GOT1 knockdown and erastin in a small panel of 188 additional PDA lines (Figure 2D) and observed that GOT1 inhibition increased 189 sensitivity to Erastin by greater than 10-fold in some cases (Figure 2C and S2B). GOT1 190 knockdown also sensitized PDA to imidazole ketone erastin (IKE), an erastin analog 191 with increased potency21 (Figure S2D). Furthermore, treating cells with a nano-molar 192 dose of erastin or IKE combined with GOT1 knockdown was cytotoxic, while single 193 treatment arms were cytostatic (Figures 2E and S2E), and these erastin doses were 194 sufficient to reduce GSH levels after 6 hours (Figure S2F). Because erastin and IKE 195 disrupt the import of cystine into the cell, we sought to inhibit cytotoxicity by co-treating 196 with exogenous cysteine sources or GSH. Indeed, supplementation with N-acetyl 197 cysteine (NAC), β-mercaptoethanol (BME), or cell permeable GSH ethyl-ester (GSH- 198 EE) prevented cell death (Figures 2F and S2G), consistent with the concept that - 22 199 cystine import through system xc is essential to maintain GSH levels . 200 201 Previous studies have found cystine levels to be limiting in the PDA microenvironment. 202 Cysteine was shown to be the second most depleted amino acid in pancreatic tumors 203 relative to adjacent healthy pancreatic tissue23 and the levels of cystine in PDA tumor 204 interstitial fluid (~50μM) which is 2-fold lower than in plasma taken from the same 205 tumor-bearing mice24. Based on these observations, we sought to test the effect of 206 GOT1 knockdown under physiological concentrations of cystine. Culturing cells in 207 tumor-relevant cystine concentrations was growth inhibitory, while cystine deprivation 208 was cytotoxic (Figures 2G and S2H). Moreover, cell viability decreased in a dose- 209 dependent manner after, and the response was potentiated by GOT1 knockdown at 210 lower cystine concentrations (Figure S2I), in agreement with our pharmacological 211 studies. These results suggest PDA require exogenous cystine for growth and cell 212 viability following GOT1 inhibition.

6 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

Figure 2. Limiting exogenous cystine causes a cytotoxic response to GOT1 inhibition. A) Screening strategy to identify metabolic dependencies induced by GOT1 knockdown. Pa-Tu-8902 iDox-shGOT1 cells were treated with vehicle or dox for 5 days followed by addition of the compounds for 72 h.

B) Log2 fold change in area under the curve (AUC) from cell viability dose response curves for each compound in the library. AUC from GOT1 knockdown conditions (dox) are relative to (mock) conditions. Sensitivity ranking (Rank) spans 1-21, signifying the most sensitive (1) to least sensitive (21) compounds in response to knockdown (n=3). C) Dose-response curves for Pa-Tu-8902 iDox-shGOT1 cells following 5 days of GOT1 knockdown and erastin for 24 hours (n=3). D) EC50 values relative to +dox for 3 iDox- shGOT1 PDA cell lines following 24 hours of erastin treatment (n=3). E) Relative Pa-Tu-8902 iDox-shGOT1 cell numbers following 5 days of GOT1 knockdown with the indicated media conditions for 24 hours. 750nM of Erastin was administered on day 1 and conditions are normalized to day 1 (n=3). F) Cell viability of Pa- Tu-8902 iDox-shGOT1 after 5 days of GOT1 knockdown then 24 hours of 750nM IKE combined with the indicated conditions. 250μM of N-acetyl-cysteine (NAC), 250μM GSH-ethyl ester (GSH-EE), and 50μM of beta-mercaptoethanol (BME) were used (n=3). G) Relative Pa-Tu-8902 iDox-shGOT1 cell numbers following 5 days of GOT1 knockdown and the indicated media conditions (n=3). 200μM of cystine is a supraphysiological concentration, while 50μM is tumor relevant. H-I) Orthotopic xenograft tumor growth from Pa-Tu-8902 iDox-shGOT1 stable cell lines co-expressing firefly luciferase (FLuc) treated with vehicle (black, n=6), dox containing food (red, n=6), cysteine-free diet (grey, n=5), or dox containing, cysteine-free food (blue, n=6). GOT1 immunoblot (I) was taken from endpoint tumors. Error bars represent mean ± SD in (B- G) or mean ± SEM in (H). Two-tailed unpaired T-test or 1-way ANOVA: Non-significant P > 0.05 (n.s. or # as noted), P ≤ 0.05 (*), ≤ 0.01 (**), ≤ 0.001 (***), ≤ 0.0001 (****). See also Figure S2. 213 To test this concept in vivo, we engrafted Pa-Tu-8902 iDox-shGOT1 cells engineered to 214 express firefly luciferase (FLuc) into the pancreas, as in Figure 1G. Tumors were

7 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

215 allowed to establish for 7 days, and treatment arms were initiated by providing dox- 216 containing food formulated with or without the non-essential amino acid cysteine. While 217 tumors in the animals fed a cysteine-free diet grew at comparable rates to tumors in 218 animals fed a control diet, dox treated tumors grew substantially slower (Figures 2H,I 219 and S2J). Mice fed with a cysteine-free diet had lower cysteine in tumors compared to 220 the control diet (Figure S2K), indicating dietary inputs can influence tumor metabolism. 221 By contrast, cysteine was not significantly altered in tumors comparing dox-single and 222 double treatment arms (Figure S2J). The difference in tumor growth or tumor burden 223 for animals on the cysteine-free diet were markedly smaller at end point, though this did 224 not reach statistical significance (Figures 2H,I). 225 226 Overall, these data indicate that PDA cultures require exogenous cystine following 227 GOT1 inhibition. We also demonstrate that this mechanism is operative in vivo, through 228 PDA tumors may acquire cysteine through alternative mechanisms when challenged by 229 chronic cysteine deprivation. 230 231 Inhibiting GSH biosynthesis potentiates the growth inhibitory effects of GOT1 232 knockdown 233 Our data indicate that PDA are heavily reliant on exogenous cystine following GOT1 234 inhibition. Previous work from our group has demonstrated that one of the downstream 235 effects of GOT1 is reducing oxidized glutathione (GSSG)8,18. Thus, we hypothesized 236 that cysteine acquisition was upregulated for GSH biosynthesis to compensate for the 237 loss of NADPH-mediated GSH salvage through the GOT1 pathway (Figure 3A). To test 238 this hypothesis, we inhibited de novo GSH biosynthesis in combination with GOT1 239 knockdown. The rate-limiting step in GSH synthesis is catalyzed by glutamate-cysteine 240 ligase (GCL), which forms gamma glutamyl-cysteine through the condensation of 241 glutamate and cysteine25. 242 243 GCL is a holoenzyme which consists of a catalytic subunit (GCLC) and a modifier 244 subunit (GCLM), and the GCLC subunit is targeted by the inhibitor buthionine 245 sulfoximine (BSO), resulting in decreased GSH production26 (Figure 3A). We observed 246 that GOT1 knockdown enhanced sensitivity to BSO after 24 hours of drug treatment, 247 contrasting an absent single agent response (Figure S3A). Exposure to BSO for 72 248 hours further augmented the sensitizing effect (Figures 3B and S3B,C). In some cases, 249 the change in EC50 was nearly 100-fold (Figures 3C and S3D), indicating potent 250 sensitization. 251 252 Recent studies have suggested that a majority of cancer cell lines survive upon 72 253 hours of BSO treatment despite potent inhibition of GSH levels at this time point. 254 Moreover, these studies have demonstrated that chronic BSO treatment, up to 9 days, 255 is required in induce cell death or stasis27. In our models, 6 hours of BSO treatment was 256 sufficient to diminish GSH levels (Figure S3E), in line with previous kinetic data27. By 257 contrast, co-treatment of GOT1 knockdown with BSO was cytotoxic at 72 hours, while 258 BSO alone was cytostatic (Figures 3D and S3F). 120 hours of treatment potentiated 259 cell death in the GOT1 knockdown condition, whereas cells regained proliferative

8 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

260 capacity under the BSO single treatment condition (Figure 3D and S3F). Cell death 261 could be prevented by supplementing exogenous GSH-EE or NAC (Figures 3E and 262 S3G), suggesting that cell death is due to perturbing GSH levels and redox balance. 263

9 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

Figure 3. Inhibiting GSH biosynthesis produces a cytotoxic response upon GOT1 knockdown. A) Schematic integrating the GSH biosynthesis and GOT1 pathways. B) Percent cell viability dose-response curves upon 72 hours of BSO treatment following 5 days of GOT1 knockdown (n=3). C) EC values 50 relative to +dox for 3 iDox-shGOT1 PDA cell lines following 72 hours of BSO treatment (n=3). D) Fold change Pa-Tu-8902 iDox-shGOT1 cell numbers following 5 days of GOT1 knockdown and treatment with the indicated conditions. 40 µM of BSO was administered on day 1. Cell numbers are normalized to day 1 for each condition (n=3). E) Percent cell viability following 72 hours of 40uM BSO or co-treatment with 0.5mM N-acetyl cysteine (NAC) or 0.5mM GSH-Ethyl Ester (GSH-EE) following 5 days of GOT1 knockdown (n=3). F) Subcutaneous xenograft growth of Pa-Tu-8902 iDox-shGOT1 cells treated with vehicle (black), 20 mg/kg BSO via drinking water (grey), doxycycline administered in the food (red), or the combination (blue). G) Relative abundance of gamma-glutamyl cysteine (γGC) from tumors in (F) (n=8). H) Relative abundances of GSH, GSSG, and the GSH/GSSG ratio from tumors in (F) (n=8). Error bars represent mean ± SD. Two-tailed unpaired T-test or 1-way ANOVA: Non-significant P > 0.05 (n.s. or # as noted), P ≤ 0.05 (*), ≤ 0.01 (**), ≤ 0.001 (***), ≤0.0001 (****). See also Figure S3.

264 Targeting GSH production with BSO is tolerated in patients, and BSO has been used in 265 combination with chemotherapy in multiple phase 1 clinical trials, (NCT00005835 and 266 NCT00002730)28,29. Thus, we sought to determine whether this combination shows 267 efficacy in vivo. We examined the effect of GOT1 and BSO in established xenograft 268 tumors. Mice were engrafted with Pa-Tu-8902 iDox-shGOT1 cells and given dox via 269 chow after 7 days. BSO was administered via drinking water on day 14. While no tumor 270 regressions were observed, the combination of GOT1 and BSO significantly slowed 271 tumor progression compared with single treatment arms (Figure 3F) and led to 272 complete stasis in one instance (Figure S3H). Knockdown was confirmed immunoblot 273 analysis (Figure S3I) and by LC-MS measurements of aspartate (Figure S3J) and on 274 whole-tumor samples. 275 276 We then measured glutathione species in tumor metabolite fractions to demonstrate the 277 pharmacodynamics of BSO. We found BSO to significantly reduce levels of gamma 278 glutamyl-cysteine, a product of GCL, which is directly inhibited by BSO25 (Figure 3G). 279 Concomitantly, we observed a significant reduction in GSH, GSSG, and the GSH/GSSG 280 ratio upon BSO treatment (Figure 3H), demonstrating BSO has on-target activity in 281 established tumors, and the tumors are under redox stress. Together, our data reveal 282 that PDA require glutathione synthesis under GOT1 deficient conditions. 283 284 GOT1 suppression sensitizes PDA to ferroptosis 285 Previous work has demonstrated that some cell types are sensitive to erastin and BSO 286 as single agents, and that these drugs can kill cells by depleting GSH. The proximal 287 effects of GSH depletion are mediated through loss of GPX4 activity, which utilizes GSH 288 as a co-factor to detoxify lipid peroxides (Figure 4A). This can lead to the lethal 289 accumulation of lipid peroxides, and ferroptosis30. Ferroptosis is a form of oxidative, 290 non-apoptotic, iron-dependent, cell death that is triggered by excessive lipid peroxide 291 levels (Figure 4A)20,31. While GOT1 inhibition does not induce ferroptosis, our data 292 suggest it may predispose PDA cells to ferroptosis. 293 294 To investigate whether GOT1 can sensitize PDA to ferroptosis, we first examined the 295 combinatorial effect of GOT1 knockdown together with RSL-3, a covalent inhibitor of

10 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

296 GPX4 and direct inducer of ferroptosis30. RSL-3 in combination with GOT1 knockdown 297 was substantially more potent than as a single agent (Figure 4B), and this effect was 298 evident across a panel of PDA lines (Figures 4C and S4A,B) and independent of dox 299 effects (Figure S4C). Pa-Tu-8902 is among the most resistant PDA cell lines to single 300 agent GPX4 inhibitors while Mia PaCa-2 was among the most sensitive (Figures 301 S4A,B), in agreement with previous reports suggesting response to GPX4 inhibitors can 302 be predicted by epithelial and mesenchymal markers32. Low concentrations of RSL-3 303 are cytostatic. When combined with GOT1 inhibition, we observed a cytotoxic response 304 (Figures 4D and S4D), suggesting GOT1 inhibition may impose GPX4 dependence in 305 a manner distinct from cell-state and in parallel with GPX4 function. 306 307 We then measured the lipid peroxide levels with the C11-BODIPY lipid peroxidation 308 sensor to investigate how inhibition of GPX4 and GOT1 affected lipid peroxidation. 309 While the effect of GOT1 inhibition on lipid ROS induction was modest GOT1 inhibition 310 combined with RSL-3 or erastin substantially upregulated lipid ROS (Figures 4E and 311 S4E) and the effect could be reversed through co-treatment with the lipophilic 312 antioxidant ferrostatin-1 (Fer-1) (Figure S4F). Next, we examined whether cell death 313 could be prevented by co-treatment with agents that relieve lipid peroxidation or chelate 314 iron20,31,33. 315 316 Co-treatments with Fer-1 and Trolox, prevented cell death induced by GOT1 317 knockdown combined with erastin, BSO or RSL-3 (Figures 4F and S4H,I), while 318 treatment with the iron chelator deferoxamine (DFO), provided substantial protection 319 (Figure 4F). To rule out the possibility that GOT1 was sensitizing PDA to alternative 320 mechanisms of cell death, namely apoptotic, necrotic, or autophagic cell death, we co- 321 treated GOT1 knockdown with well-characterized inhibitors of these cell death 322 pathways. Indeed, the addition of a pan-caspase inhibitor (Z-VAD-FMK), RIPK-1 323 inhibitor (Necrostatin-1), or lysosomal acidification inhibitor (Bafilomycin A1) offered 324 limited protection from cell death compared with lipophilic antioxidants or iron chelation 325 (Figures 4G and S4H,I), suggesting ferroptosis is the predominant mechanism of cell 326 death. Next, we explored triggering ferroptosis by (–)–FINO2 which causes iron 327 oxidation and indirectly inhibits GPX4 activity34. Indeed, GOT1 suppression sensitized 328 PDA to (–)–FINO2 in an additive manner (Figure S4J). Overall, our data demonstrate 329 GOT1 inhibition primes PDA for ferroptosis. 330 331 GOT1 inhibition primes PDA for ferroptosis by promoting labile iron release in 332 response to metabolic stress 333 Ferroptosis is driven by oxidation of polyunsaturated fatty acids in the cell membrane, a 334 process that is catalyzed by iron. Thus, ferroptosis is coupled to the cell’s metabolic 335 state31. For example, cells can be primed for ferroptosis by enriching polyunsaturated 336 acid composition in cell the membrane in a HIF-2α dependent manner, by depleting 337 essential co-factors for GPX4-30 or ferroptosis suppressor protein 1 (FSP1)- 338 mediated35,36 lipid antioxidant activity, or by increasing intracellular free iron levels37. 339 Because our data indicated that GOT1 inhibition sensitizes PDA to ferroptosis in a 340 manner that is independent of epithelial or mesenchymal cell state, we sought to

11 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

341

12 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

Figure 4. GOT1 suppression sensitizes PDA to ferroptosis. A) Schematic of the GPX4 arm of ferroptosis. B) Cell viability dose-response curves upon 24 hours of RSL-3 treatment following 5 days of GOT1 knockdown (n=3). Viability is normalized to a 0.1% DMSO vehicle control. C) EC50 values relative to +dox for 5 iDox-shGOT1 PDA cell lines following 24 hours of RSL-3 treatment (n=3). D) Fold change Pa-Tu- 8902 iDox-shGOT1 cell numbers following 5 days of GOT1 knockdown and 24 hours treatment with the indicated conditions. 32nM of RSL-3 was administered on day 1. Cell numbers are normalized to day 1 for each condition (n=3). -/+ dox conditions (grey and light grey) contained a 0.1% DMSO vehicle control. E) Fold change in viable MIA PaCa-2 iDox-shGOT1 cells positive for C-11 BODIPY, a dye sensitive to lipid ROS, following 5 days of GOT1 knockdown. Cells were treated with the indicated conditions for 6 hours prior to measurements: vehicle (0.1% DMSO) -/+ dox (black and grey), 1μM RSL-3. Data are normalized to the –dox and vehicle-treated condition (n=2). F) Cell viability of Pa-Tu-8902 iDox-shGOT1 cultured in vehicle (0.1% DMSO) -/+ dox (black and light grey), drug (32nM RSL-3, 750nM Erastin, 40μM BSO) -/+ doxycycline (grey and red), or drug and dox (blue) in the presence of lipophilic antioxidants 1μM Fer-1 and 100μM Trolox, or an iron chelator 10μM DFO (deferoxamine). Viability was assessed after 24 hours of treatment for RSL-3 and Erastin conditions and 72 hours for BSO treatment conditions. GOT1 was knocked down for 5 days prior to treatment. Data are normalized to the -dox and vehicle treated control (n=3) G) Cell viability of Pa-Tu-8902 iDox-shGOT1 cultured in vehicle (0.1% DMSO) -/+ dox (black and light grey), drug (32nM RSL-3, 750nM Erastin, 40μM BSO) -/+ dox (grey and red), or drug and dox (blue) in the presence of 10μM Necrostatin-1 (Nec-s, apoptosis inhibitor), 50 μM ZVAD-FMK (Z-Vad, apoptosis inhibitor), or 1 nM bafilomycin A1 (BA-1, autophagy inhibitor). Viability was assessed after 24 hours of treatment for RSL-3 and Erastin conditions and 72 hours for BSO treatment conditions. GOT1 was knocked down for 5 days prior to treatment. Data are normalized to the -dox and vehicle treated control (n=3). Error bars represent mean ± SD. Two-tailed unpaired T-test or 1-way ANOVA: Non-significant P > 0.05 (n.s. or # as noted), p ≤ 0.05 (*), ≤ 0.01 (**), ≤ 0.001 (***), ≤0.0001 (****). See also Figure S4.

342 identify metabolic features that would promote ferroptosis sensitivity in addition to 343 modulating NADPH and GSH levels (Figure S5A). 344 345 First, we examined the possibility that GOT1 could be altering phospholipid 346 composition. HIF-2α stabilization has been recently shown to promote a ferroptosis 347 primed cell-state by selectively enriching for polyunsaturated acids in clear-cell 348 carcinomas38. This vulnerability was mediated by hypoxia-inducible lipid droplet- 349 associated protein (HILPDA). In our PDA models, both HIF-2α levels and HILPDA 350 expression were unchanged in response to GOT1 knockdown (Figures S5B-D). 351 Next, we examined the possibility that GOT1 inhibition was priming cells to ferroptosis 352 by increasing intracellular iron levels. First, we tested whether GOT1 inhibition would 353 sensitize PDA to perturbations in intracellular iron. We began by testing whether GOT1 354 knockdown rendered PDA susceptible to increased iron loads using ferric ammonium 355 citrate (FAC) (Figures 5A and S5E). This is consistent with the idea that GOT1 356 inhibition promotes an oxidized state by depleting both NADPH and GSH while 357 increasing intracellular ROS8,18. This oxidized metabolic state primes cells for iron- 358 mediated lipid peroxidation. 359 360 We then used Calcein-AM dye, a fluorescein-derived probe that is quenched when 361 bound to ferrous iron (Fe2+)39, to directly assess how GOT1 knockdown was impacting 362 cellular iron levels. Calcein-AM staining of GOT1 proficient cells defined basal 363 fluorescence. Interestingly, GOT1 knockdown cells shifted fluorescence distribution to 364

13 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

365

14 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

Figure 5. GOT1 inhibition promotes autophagic labile iron release. A) Cell viability dose-response curves upon 72 hours of FAC treatment following 5 days of GOT1 knockdown (n=3). Viability is normalized to a 0.1% DMSO vehicle control. B-D) Calcein-AM histogram (B), mean fluorescence intensity (MFI) (C), and relative labile iron (D) in Pa-Tu-8902 iDox-shGOT1 cells after 5 days of GOT1 knockdown (n=4). E) Mechanisms regulating intracellular iron levels and the Fenton Reaction. F) Immunoblot analysis of autophagy markers NCOA4, LC-3 A/B, and P62 in Pa-Tu-8902 and MIA PaCa-2 iDox-shGOT1 cells following 5 days of knockdown. G) Quantification of P62 from (F) normalized to the GAPDH loading control. H) Quantification of autophagic flux in MT3 cells treated with vehicle or PF-04859989 (n=3) frames. I) Cell viability of Pa-Tu-8902 iDox-shGOT1 after 24 hour treatments with 10 μM (–)–FINO2 as a single agent, or combined with dox, or dox plus 10 μM deferoxamine (DFO), 8nM bafilomycin A1 (BA-1), or 1.5 μM Hydroxychloroquine (HCQ) (n=3). J) Mitostress assay measuring OCR in Mia PaCa-2 iDox-shGOT1 cells (n=3). K) AMP/ATP ratio measured by liquid-chromatography tandem mass spectrometry, normalized to - dox (n=4). L) GOT1 inhibition promotes the release of labile iron through autophagy in response to metabolic stress and sensitizes PDA to ferroptosis. Error bars represent mean ± SD. Two-tailed unpaired T- test or 1-way ANOVA: Non-significant P > 0.05 (n.s. or # as noted), P ≤ 0.05 (*), ≤ 0.01 (**), ≤ 0.001 (***), ≤0.0001 (****). See also Figure S5.

366 lower intensity, indicating labile iron pools were increased following GOT1 knockdown 367 (Figures 5B-D and S5F,G). Iron levels can be altered by downregulating iron efflux, 368 upregulating iron uptake, or promoting the degradation of intracellular iron carriers— 369 ferritin or heme (Figure 5E). While previous studies suggest 2017), GOT1 knockdown 370 did not upregulate expression of iron transport proteins (SLC40A1 and TFRC) (Figures 371 S5H,I). Moreover, expression of heme oxygenase 1 (HMOX1), which releases labile 372 iron through the degradation of heme was unaltered in Pa-Tu-8902. It was however 373 modestly upregulated in Mia PaCa-2 (Figure S5I), suggesting this mechanism may 374 contribute to increasing intracellular iron pools in some PDA specimens. 375 376 Autophagy is a catabolic process that protects cells from metabolic stress induced by 377 nutrient deprivation40. Moreover, autophagy is heavily utilized by pancreatic 378 cancers6,7,41,42. Indeed, GOT1 knockdown upregulated autophagic flux, indicated by the 379 increase in LC3-B (Figure 5F), consistent with previous work in osteosarcoma cells43. 380 Moreover, GOT1 inhibition led to decreased p62, which is selectively degraded during 381 autophagy (Figure 5F,G), and promoted the enrichment of autolysosomes (Figures 5H 382 and S5J). Because GOT1 inhibition did not upregulate the expression of iron importers 383 or exporters to regulate intracellular iron (Figures S5H,I), we wondered if cells may be 384 liberating ferritin-bound iron. transferrin import is required for ferroptosis44 and 385 upregulation of transferrin through the iron starvation response can promote 386 ferroptosis37. 387 388 The majority of stored iron is bound by ferritin and can be released via NCOA4- 389 dependent autophagy, in a process termed ferritinophagy16. NCOA4 is autophagosome 390 cargo receptor that binds to the ferritin heavy chain sequestering ferritin for degradation 391 by the autolysosome to release labile iron16. Upregulated autophagic flux (Figure 5F-H) 392 was matched by NCOA4 expression (Figure 5F), suggesting the changes in labile iron 393 pools may occur through ferritinophagy. Ferritinophagy can indirectly promote 394 ferroptosis by releasing free iron required for lipid peroxidation. Moreover, knockdown of 395 NCOA4 inhibits ferroptosis in HT-1080 fibrosarcoma and in Panc-1, a PDA cell line, 396 potentially by reducing labile iron pools45,46. Importantly, supplementing iron chelators

15 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

397 (DFO) or inhibitors of lysosomal acidification, [bafilomycin A1 (BA-1) or 398 Hydroxychloroquine (HCQ)] prevented the additive effect of GOT1 inhibition and (–)– 399 FINO2 (Figure 5G). Because (–)–FINO2 oxidizes iron and since GOT1 inhibition 400 liberates iron, these data support the model that GOT1 promotes labile iron through 401 ferritinophagy. 402 403 Autophagy is regulated by cellular energetic and nutrient status. Accordingly, we 404 hypothesized that GOT1 inhibition may induce autophagy in response to metabolic 405 stress. First, using bioenergetic profiling with the Seahorse instrument, we found that 406 GOT1 knockdown lowered basal, maximal, and spare respiratory capacities (Figure 407 5J), indicating decreased mitochondrial fitness. Further, using LC/MS we found that 408 PDA were experiencing energetic stress, marked by an upregulated AMP/ATP ratio 409 (Figures 5K and S5K). Iron uptake is regulated to support adaption to high oxygen 410 conditions37 and to support OxPHOS by supplying metabolically active iron for iron- 411 sulfur cluster and heme biogenesis47. Thus, PDA cells may liberate iron as support 412 mitochondrial metabolism following GOT1 loss; however, this results at the cost of 413 rendering PDA susceptible to oxidative assault. Overall, our data suggest that GOT1 414 primes ferroptosis by promoting the autophagic release of iron in response to metabolic 415 stress (Figure 5L). 416 417 Discussion 418 In this study we report that the non-canonical malate-aspartate shuttle function is a key 419 contributor to the metabolic fidelity of PDA cells by maintaining NADPH pools and 420 mitochondrial anaplerosis. Inhibition of GOT1 suppresses the growth of numerous PDA 421 cell lines, primary culture models, and xenograft tumors, while rendering cells 422 susceptible to ferroptosis. Ferroptosis could be triggered by inhibition cystine import, 423 glutathione synthesis, or GPX4 in synergy with GOT1 (Figures 2-4), which we ascribe 424 to the promotion of intracellular iron levels through the degradation of ferritin along with 425 the suppression of NADPH. 426 427 The dependency on exogenous cystine in GOT1 knockdown cells was identified using a 428 synthetic lethal chemical screening strategy (Figures 2A-C and S2A-B). We then 429 demonstrated that exogenous cystine through system xC- enables de novo GSH 430 biosynthesis, which is required to compensate for decreased GSH availability. When 431 GOT1 is inhibited, NADPH availability is decreased and GSH cannot be sufficiently 432 regenerated by reducing GSSG. This metabolic rewiring could be exploited by dietary 433 means, which has a major influence on the nutrient composition within pancreatic 434 tumors24. This result adds to a growing body of literature indicating how the metabolic 435 environment can influence sensitivity to therapy48,49. 436 437 The inhibition of de novo GSH biosynthesis with BSO also potentiated tumor inhibition 438 by GOT1. BSO has been used in clinical trials and is tolerated in patients28,29. Previous 439 efforts with BSO have indicated that many tumor types employ compensatory 440 mechanisms to tolerate glutathione inhibition27,50. Indeed, we too have observed that 441 inhibition of de novo glutathione biosynthesis is not sufficient to induce ferroptosis in 442 PDA. This points to GOT1 inhibition as a logical combination therapy to enhance

16 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

443 therapeutic efficacy of BSO. To this end, we and others engaged in drug discovery 444 campaigns to develop small inhibitors of GOT110,11,12,13. Among these, we found that the 445 KATII inhibitor PF-04859989 also has potent GOT1 inhibitory activity13, and we apply 446 that herein as a tool compound. Future studies are yet required to improve the drug like 447 properties of this molecule for in vivo studies. 448 449 Reduced glutathione is a co-factor for GPX4 that is used to maintain lipid antioxidant 450 function. It is also conceivable that the non-canonical malate-aspartate shuttle may 451 supply NADPH to FSP1, which reduces the co-factor CoQ10 to inhibit lipid peroxides in 452 parallel with GPX435,36. This could also account for how GOT1 inhibition sensitizes PDA 453 to ferroptosis (Figures 3-5). These results would be consistent with previous studies 454 demonstrating GOT1 inhibition can induce ROS8,12 and radiosensitize PDA both in vitro 455 and in vivo18. Future studies will be required to examine the role of FSP1 and potential 456 interactions with the GOT1 pathway in PDA. 457 458 In line with the name, ferrous iron is required for ferroptosis by contributing to the 459 oxidation of membrane PUFAs, either as free iron or as a co-factor for lipoxygenase 460 enzymes. Iron levels can be altered by upregulating iron import, downregulating iron 461 export, or degrading iron carriers. Our studies indicate that GOT1 knockdown promotes 462 the release of intracellular iron, and sensitized PDA to the deleterious effects of iron 463 oxidation and iron loading. GOT1 knockdown did not alter the expression of iron uptake 464 or efflux proteins. Rather, GOT1 knockdown leads to a catabolic state marked by 465 diminished mitochondrial activity and elevated AMP/ATP. To support metabolic 466 homeostasis, GOT1 knockdown cells activate autophagy, a process that enables cells 467 to degrade cellular components through the lysosome and recycle these to adapt to 468 nutrient stress. Concurrently, autophagy in these NCOA4 expressing PDA cells also 469 leads to the turnover of ferritin-bound iron via ferritinophagy16. This leads to higher 470 levels of free iron, and by extension, promotes susceptibility to ferroptosis. It is not 471 entirely clear why PDA cells would release free iron through ferritinophagy upon GOT1 472 knockdown. It has been proposed by that labile iron, which is metabolically active and 473 can be utilized by iron-containing enzymes, is released to support DNA synthesis, 474 epigenetic modification, and the biogenesis of iron-sulfur clusters to support 475 mitochondrial metabolism47. Our data are consistent with a model whereby cells can 476 employ ferritinophagy to release iron for metabolic demands, but with the cost of 477 rendering cells susceptible to ferroptosis. 478 479 The role of GOT1 in ferroptosis has been the subject of previous study in several other 480 tumor types. Our data lie in contrast to some previous work, which have suggested that 481 GOT1 inhibition by genetic or pharmacological means in other tumor types protects 482 cells from ferroptosis by blocking mitochondrial metabolism20,45,51. Recent studies 483 suggest the genotype32, nutrient environment52, tissue of origin30,38, and cell- 484 autonomous metabolism35-37 are major drivers of ferroptosis sensitivity. The differences 485 emerging from these studies likely reflect the incomplete understanding of how these 486 various factors dictate sensitivity to ferroptosis. Thus, our study provides clarity 487 regarding the metabolic regulation of ferroptosis in PDA. 488

17 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

489 Significance 490 PDA is a notoriously deadly disease with a ten percent 5-year survival owing in largely 491 to a lack of effective therapeutic options. PDA cells rewire metabolism to survive and 492 proliferate with a nutrient-deprived and metabolically harsh environment. This study not 493 only demonstrates that the GOT1 non-canonical malate-aspartate pathway is essential 494 for maintaining biosynthetic and antioxidant fidelity in pancreatic cancer, it also 495 illustrates that promoting a catabolic, oxidative cell state leads to the autophagy- 496 dependent release of labile iron. This sensitizes pancreatic cancer to ferroptosis by 497 pharmacological or dietary means. These applications represent alternative contexts for 498 the repurposing clinical agents and could reveal novel therapeutic strategies that 499 selectively exploit the unique metabolic demands of pancreatic tumors. 500 501 502 KEY RESOURCES TABLE REAGENT or RESOURCE SOURCE IDENTIFIER Antibodies Rabbit monoclonal anti-GOT1 Abcam Cat#: ab171939 Anti-Ki67 CST Cat#: 9027 Rabbit monoclonal 5A1E anti-Cleaved Caspase 3 CST Cat#:9664 Anti-HSP90 CST Cat#: 4877 Rabbit monoclonal E1E9V anti-Vinculin CST Cat#:13901 Anti-Rabbit IgG HRP CST Cat#: 7074 Anti-Rabbit GAPDH CST Cat#: 2118 Anti-Rabbit LC3A/B CST Cat#: 12741 Anti-Rabbit NCOA4 (ARA70) Bethyl Laboratories A302-272A Anti-Rabbit HIF-2 Abcam Cat#: ab131743 Bacterial and Virus Strains Lentiviral plasmid: Tet-pLKO-puro A gift from Dmitri Addgene, 21915 Wiederschain Tet-pLKO-shGOT1 coding region (shGOT1 #1) Son et al. 2013 TRCN0000034784 Tet-pLKO-shGOT1 3’UTR (shGOT1 #3) Nelson et al. 2019 N/A Tet-pLKO-sh non-targeting (shNT) Nelson et al. 2019 N/A pFUGW-firefly luciferase (FL) Smith et al. 2004 Addgene, 14883 Chemicals, Peptides, and Recombinant Proteins Doxycycline Hyclate Sigma Aldrich Cat#:D9891 RIPA buffer Sigma Aldrich Cat#:R0278 L-Ascorbic acid Sigma-Aldrich Cat#: A5960; CAS: 50-81-7 2-AAPA hydrate Sigma-Aldrich Cat#: A4111 6-Aminonicotinamide (6-AN) Cayman Chemical Cat#: 10009315; CAS: 329-89-5 FX-11 Sigma-Aldrich Cat#: 427218; CAS: 213971-34-7 Paclitaxel Cayman Chemical Cat#: 10461; CAS: 33069-62-4 Gemcitabine Cayman Chemical Cat#: 11690, CAS: 95058-81-4 Cisplatin Cayman Chemical Cat#: 13119, CAS: 15663-27-1

18 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

Lovastatin Cayman Chemical Cat#: 10010338, CAS: 75330-75-5 Veliparib (ABT-888) Cayman Chemical Cat#: 11505, CAS: 912445-05-7 Phenformin (hydrochloride) Cayman Chemical Cat#: 14997, CAS: 834-28-6 Simvastatin Cayman Chemical Cat#: 10010344, CAS: 79902-63-9 TOFA Cayman Chemical Cat# 10005263, CAS: 54857-86-2 FK-866 Cayman Chemical Cat#: 13287, CAS: 658084-64-1 5-FU Cayman Chemical Cat#: 14416, CAS: 51-21-8 2-deoxy-D-Glucose Cayman Chemical Cat#: 14325, CAS: 154-17-6 Auranofin Cayman Chemical Cat#: 15316, CAS: 34031-32-8 Methotrexate Cayman Chemical Cat#: 13960, CAS: 59-05-2 Koningic Acid Cayman Chemical Cat#: 14079, CAS: 57710-57-3 Erastin Cayman Chemical Cat#: 17754, CAS: 571203-78-6 Oligomycin A Cayman Chemical Cat#: 11342, CAS: 579-13-5 Bafilomycin A1 Cayman Chemical Cat#: 11038, CAS: 88899-55-2 Deoxynyboquinone (DNQ) A gift from David A. PubChem CID: Boothman 295934 Imidazole ketone erastin (IKE) MedChemExpress Cat#: Y-114481 , CAS: 1801530-11-9 N-Acetyl-L-cysteine (NAC) Sigma-Aldrich Cat#: A8199; CAS: 616-91-1 Glutathione ethyl ester (GSH-EE) Cayman Chemical Cat#: 14953, CAS: 92614-59-0 2-Mercaptoethanol (BME) Sigma-Aldrich Cat#: M6250, CAS: 60-24-2 L-Cystine Sigma-Aldrich Cat#: C8755, CAS: 56-89-3 L-Buthionine-(S,R)-Sulfoximine (BSO) Cayman Chemical Cat#: 14484, CAS: 83730-53-4 (1S,3R)-RSL3 Cayman Chemical Cat#: 19288, CAS: 1219810-16-8 Ferrostatin-1 Cayman Chemical Cat#: 17729, CAS: 347174-05-4 Trolox Cayman Chemical Cat#: 10011659, CAS: 53188-07-1 Deferoxamine (mesylate) (DFO) Cayman Chemical Cat#: 14595, CAS: 138-14-7 PF-04859989 hydrochloride Cayman Chemical Cat#: PZ0250, CAS: 177943-33-8 Necrostatin-1 Cayman Chemical Cat#: 11658, CAS: 4311-88-0

19 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

Z-VAD(OMe)-FMK Cayman Chemical Cat#: 14463, CAS: 187389-52-2 FG-4592 Cayman Chemical Cat#: 15294, CAS: 808118-40-3 FINO2 Cayman Chemical Cat#: 25096, CAS: 869298-31-7 Ferric ammonium citrate Sigma-Aldrich Cat#: F5879, CAS: 1185-57-5 BODIPY™ 581/591 C11 (Lipid Peroxidation Sensor) Invitrogen Cat#: D3861 Calcein, AM Invitrogen Cat#: C3099 Critical Commercial Assays Cell Proliferation Reagent WST-1 Sigma Cat#: 11644807001 Cell Titer Glo 2.0 Promega Cat#:G9241 Cyquant ThermoFisher Cat#:C7026 Beetle Luciferin Promega Cat#:E1605 MycoAlert Lonza Cat#:LT07-318 GSH-Glo Promega Cat#:V6911 iScript cDNA synthesis kit BioRad Cat#:1708890 Experimental Models: Cell Lines Human: PL45 control This paper N/A Human: PL45 shNT This paper N/A Human: PL45 iDox-shGOT1 #1 This paper N/A Human: PL45 iDox-shGOT1 #3 This paper N/A Human: Capan-1 control This paper N/A Human: Capan-1 shNT This paper N/A Human: Capan-1 iDox-shGOT1 #1 This paper N/A Human: Capan-1 iDox-shGOT1 #3 This paper N/A Human: Pa-Tu-8902 control This paper N/A Human: Pa-Tu-8902 shNT This paper N/A Human: Pa-Tu-8902 iDox-shGOT1 #1 This paper N/A Human: Pa-Tu-8902 iDox-shGOT1 #3 This paper N/A Human: Pa-Tu-8902 iDox-shGOT1 #1, pFUGW-Firefly This paper N/A Luciferase Human: BxPC-3 control This paper N/A Human: BxPC-3 shNT This paper N/A Human: BxPC-3 iDox-shGOT1 #1 This paper N/A Human: BxPC-3 iDox-shGOT1 #3 This paper N/A Human: Pa-Tu-8988T control This paper N/A Human: Pa-Tu-8988T shNT This paper N/A Human: Pa-Tu-8988T iDox-shGOT1 #1 This paper N/A Human: Pa-Tu-8988T iDox-shGOT1 #3 This paper N/A Human: MIA PaCa-2 control This paper N/A Human: MIA PaCa-2 shNT This paper N/A Human: MIA PaCa-2 iDox-shGOT1 #1 This paper N/A Human: MIA PaCa-2 iDox-shGOT1 #3 This paper N/A

20 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

Human: Panc10.05 control This paper N/A Human: Panc10.05 shNT This paper N/A Human: Panc10.05 iDox-shGOT1 #1 This paper N/A Human: Panc10.05 iDox-shGOT1 #3 This paper N/A Human: Panc03.27 control This paper N/A Human: Panc03.27 shNT This paper N/A Human: Panc03.27 iDox-shGOT1 #1 This paper N/A Human: Panc03.27 iDox-shGOT1 #3 This paper N/A Human: YAPC control This paper N/A Human: YAPC shNT This paper N/A Human: YAPC iDox-shGOT1 #1 This paper N/A Human: YAPC iDox-shGOT1 #3 This paper N/A Human: HupT3 control This paper N/A Human: HupT3 shNT This paper N/A Human: HupT3 iDox-shGOT1 #1 This paper N/A Human: HupT3 iDox-shGOT1 #3 This paper N/A Human: Panc-1 control This paper N/A Human: Panc-1 shNT This paper N/A Human: Panc-1 iDox-shGOT1 #1 This paper N/A Human: Panc-1 iDox-shGOT1 #3 This paper N/A Human: Capan-2 control This paper N/A Human: Capan-2 shNT This paper N/A Human: Capan-2 iDox-shGOT1 #1 This paper N/A Human: Capan-2 iDox-shGOT1 #3 This paper N/A Human: UM147 control This paper N/A Human: UM147 shNT This paper N/A Human: UM147 iDox-shGOT1 #1 This paper N/A Human: UM147 iDox-shGOT1 #3 This paper N/A Human: UM5 control This paper N/A Human: UM5 shNT This paper N/A Human: UM5 iDox-shGOT1 #1 This paper N/A Human: UM5 iDox-shGOT1 #3 This paper N/A Human: UM90 control This paper N/A Human: UM90 shNT This paper N/A Human: UM90 iDox-shGOT1 #1 This paper N/A Human: UM90 iDox-shGOT1 #3 This paper N/A Human: UM19 control This paper N/A Human: UM19 shNT This paper N/A Human: UM19 iDox-shGOT1 #1 This paper N/A Human: UM19 iDox-shGOT1 #3 This paper N/A Human: HPNE(V) iDox-shGOT1 #1 This paper N/A

21 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

Human: HPNE(V) iDox-shGOT1 #3 This paper N/A Human: IMR-90 iDox-shGOT1 #1 This paper N/A Human: IMR-90 iDox-shGOT1 #3 This paper N/A

Human: hPSC iDox-shGOT1 #1 γ This paper N/A Human: hPSC iDox-shGOT1 #3 This paper N/A Murine: KPC-MT3 LC3-GFP-RFP This paper N/A Oligonucleotides Human: GOT1 Fw_caactgggattgacccaact This paper N/A Human: GOT1 Rev_ggaacagaaaccggtgctt This paper N/A Human: HMOX1 Fw_ggcagagggtgatagaagagg This paper N/A Human: HMOX1 Rev_agctcctgcaactcctcaaa This paper N/A Human: TFRC Fw_acctgtccagacaatctccag This paper N/A Human: TFRC Rev_tgttttccagtcagagggaca This paper N/A Human: SLC40A1 Fw_ccaaagggattggattgttg This paper N/A Human: SLC40A1 Rev_ccttcgtattgtggcattca This paper N/A Human: HILPDA Fw_ aagcatgtgttgaacctctacc This paper N/A Human: HILPDA Rev_ tgtgttggctagttggcttct This paper N/A pBABE-puro mCherry-EGFP-LC3B Addgene 22418 Software and Algorithms Morpheus Broad Institute https://software.broa dinstitute.org/morph eus/ ImageJ ImageJ https://imagej.nih.go v/ij/ Prism 7 GraphPad https://www.graphpa d.com/scientific- software/prism/ FlowJo v.10 FlowJo https://www.flowjo.co m/ 503 504 Contact for Reagent and Resource Sharing 505 Further information and requests for resources and reagents should be directed to and will be 506 fulfilled by the Lead Contact, Costas A. Lyssiotis ([email protected]). 507 508 Cell Culture 509 PL45, Capan-1, BxPC-3, MIA PaCa-2, Panc10.05, Panc03.27, PANC-1, Capan-2, HPNE (V), 510 IMR-90 were obtained from ATCC. Pa-Tu-8902, Pa-Tu-8988T, YAPC, and Hup T3 were 511 obtained from DSMZ. Human pancreatic stellate cells (hPSC) were a generous gift form Rosa 512 Hwang (Hwang et al., 2008). The UM PDA primary cell cultures (UM147, UM5, UM90, and 513 UM19) were obtained from surgically-resected samples and established through murine 514 xenograft (Li et al., 2007). KPC-MT3 murine PDA cell lines were a generous gift from Dr. David 515 Tuveson. All commercial cell lines and UM PDA primary cultures were validated by STR 516 profiling and tested negative for mycoplasma infection (Lonza, LT07-701). Cells were 517 maintained under standard conditions at 37ºC and 5% CO2. Cells were grown either in regular 518 DMEM (GIBCO, #11965) or RPMI (GIBCO, #11875), or in DMEM without cystine (GIBCO, 519 #21013024) or RPMI (GIBCO, A1049101) supplemented with 10% FBS (Corning, 35-010-CV) 520 unless otherwise indicated. Cultures involving inducible short-hairpin mediated knockdown were

22 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

521 supplemented with doxycycline-hyclate (Dox) at 1µM/mL (Sigma, D9891) for 5 days prior to 522 experiments. 523 524 KPC-MT3 LC3 tandem GFP-RFP cells were established through transfection with pBABE-puro 525 mCherry-EGFP-LC3B (Addgene plasmid #22418) followed by selection and single cell cloning. 526 For autophagic flux quantification, cells were seeded into 4 well chamber slides (Corning, 527 354104) and fixed in 4% paraformaldehyde (ThermoFisher, 28908) 24 hours following seeding. 528 Coverslips were mounted in DAPI containing mounting solution (Life Technologies P36931). 529 Cells were imaged using a Nikon A1 Confocal in FITC, RFP and DAPI channels. The ratio of 530 red:yellow puncta was determined by counting puncta using the particle analysis in imageJ. 531 532 Lentiviral-mediated shRNA Transduction 533 Parental PDA cell lines were transduced with lentivirus containing short hairpin RNA plasmids at 534 optimized viral titers. Stable cell lines were established post-puromycin selection. 535 536 Clonogenic assays 537 Cells were plated in a 6-well plate in biological triplicates at 300-600 cells per well in 2mL of 538 media. Dox-media were changed every 2 days. Assays were concluded after 10-15 days by 539 fixing in -20oC cold 100% methanol 10 min and staining with 0.5% crystal violet 20% methanol 540 solution for 15 min. Colonies were quantified using ImageJ or manually counted. 541 542 Cell proliferation assays 543 Cells were seeded in a 96-well plate at 1,000 cells per well in 0.1mL of media. Indicated 544 treatments were applied the subsequent day. Media was changed every 2 days. At the indicated 545 time points, media was aspirated and frozen. 100 µL of CyQUANT (Invitrogen, C7026) 546 to each well for measurements. 10µL of WST-1 reagent directly to the culture media (Sigma, 547 #11644807001). Relative proliferation was determined by the fluorescence intensity at 530nm 548 for CyQuant or 450nm for WST-1 using a SpectraMax M3 plate reader. 549 550 Cell viability assays 551 Cells were plated in a 96- or 384-well plate format at 1,000 cells per well Cells were allowed to 552 seed overnight, then treated with compounds at indicated concentrations and for indicated 553 lengths of time. All viability assays utilized the Cell-Titer-Glo 2.0 reagent (Promega, G9243) 554 according to the manufacturer’s instructions. Media was aspirated followed by the addition of 555 100 µL of Cell-Titer-Glo 2.0 reagent to each experimental well. Plates were gently agitated for 556 10 minutes to promote adequate mixing. Luminescence was subsequently measured using a 557 SpectraMax M3 plate reader. 558 559 Quantitative RT-PCR 560 Total RNA was extracted using the RNeasy Mini Kit (Qiagen, 74104) and reverse 561 transcription was performed from 2 µg of total RNA using the iScript cDNA synthesis kit 562 (BioRad,1708890) according to the manufacturer’s instructions. Quantitative RT–PCR was 563 performed with Power SYBR Green dye (Thermo, 4367659) using a QuantStudio 3 System 564 (Thermo). PCR reactions were performed in triplicate and the relative amount of cDNA was 565 calculated by the comparative CT method using an RPS21 as an endogenous control. RT- 566 PCR was performed in a least 3 biological replicates. 567 568 Detection of Reactive Oxygen and Labile Iron by Flow cytometry 569 Cells were plated in 6-well plates two days before incubation with indicated treatments. Cells 570 were then washed twice with 1x PBS, and stained for 20-30 (Invitrogen, C1430) minutes with 571 2µM C11-BODIPY (Invitrogen, D3861) or for 10 minutes with 0.2µM Calcein-AM (Invitrogen,

23 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

572 C1430) in phenol red-free DMEM. Cells were co-stained with Sytox-blue (Invitrogen, S34857) to 573 account for cell viability. Following staining, cells were washed twice with PBS, trypsinized 574 (0.25%, Life Technologies, 25200-056), and naturalized with pure FBS at a 1:1 volume. Cells 575 were then collected in 500uL PBS, and moved to round bottom 96-well plates, on ice, for 576 measurements. A minimum of 8000 cells were analyzed per condition. C11-BODIPY and 577 Calcein-AM signals were analyzed in the FITC channel, while Sytox-blue was analyzed in the 578 DAPI channel on a ZE5 Cell analyzer (Bio-Rad). Analysis of data was performed using FlowJo 579 v.10 software. See Supplementary Figures 6 and 7 for representative gating. Relative labile 580 iron levels were calculated based on the ratio of Calcein-AM mean fluorescence intensity (MFI) 581 of control vs. dox-treated samples. 582 583 Xenograft Studies 584 Animal experiments were conducted in accordance with the Office of Laboratory Animal Welfare 585 and approved by the Institutional Animal Care and Use Committees of the University of 586 Michigan. NOD scid gamma (NSG) mice (Jackson Laboratory, 005557), 6-8 or 8-10 weeks old 587 of both sexes, were maintained in the facilities of the Unit for Laboratory Animal Medicine 588 (ULAM) under specific pathogen-free conditions. Stable PDA cell lines containing a dox- 589 inducible shRNA against GOT1 were trypsinzied and suspended at 1:1 ratio of DMEM (Gibco, 590 11965-092) cell suspension to Matrigel (Corning, 354234). 150-200 μL were used per injection. 591 For subcutaneous xenograft studies, 0.5x106 cells were implanted into the lower flanks. 592 Doxycycline (dox) chow (BioServ, F3949) was fed to the +dox groups. Orthotopic tumors were 593 established by injecting 5x104 Pa-Tu-8902 iDox-shGOT1 #1 pFUGW-Firefly Luciferase into 8-10 594 week old NSG mice. Cysteine-free chow (LabDiet) was customized from Baker Amino Acid 595 (LabDiet, 5CC7) to remove cysteine and balance protein levels with increased valine and 596 aspartic acid. BSO was delivered in the drinking water at 20 mM. All treatments began on day 7 597 after implantation. 598 599 Subcutaneous tumor size was measured with digital calipers at the indicated endpoints. Tumor 600 volume (V) was calculated as V = 1/2(length x width2). Bioluminescence (BLI) of orthotopic 601 tumors were measured via IVIS SpectrumCT (PerkinElmer) following an intraperitoneal injection 602 of 100 μL beetle luciferin (40 mg/mL in PBS stock) (Promega, E1605). BLI was analyzed with 603 Living Image software (PerkinElmer). At endpoint, final tumor volume and mass were measured 604 prior to processing. Tissue was either fixed in zinc formalin fixative (Z-fix, Anatech LTD, #174) 605 for >24 hours for histological and/or histochemical analysis, or snap-frozen in liquid nitrogen 606 then stored at -80°C until metabolite or protein analysis. 607 608 Western blot analysis 609 Stable shNT and shGOT1 cells were cultured with or without dox media and protein lysates 610 were collected after five days using RIPA buffer (Sigma, R0278) containing protease inhibitor 611 cocktail (Sigma/Roche, 04 693 132 001). Samples were quantified with Pierce BCA Protein 612 Assay Kit (ThermoFisher, 23225). 10 to 40 µg of protein per sample were resolved on NuPAGE 613 Bis-Tris Gels (Invitrogen, NP0336) and transferred to a Immobilon-FL PVDF membrane 614 (Millipore, IPVH00010). Membranes were blocked in 5% non-fat dry milk in distilled H2O prior to 615 incubation with the primary antibody. The membranes were washed with TBS-Tween followed 616 by a 1h exposure to the appropriate horseradish peroxidase-conjugated secondary antibody. 617 The membranes were washed in de-ionized water for 15-30 minutes then visualized using a 618 Bio-Rad ChemiDox MP Imaging System (Bio-Rad, 17001402). The following antibodies were 619 used: anti-aspartate aminotransferase (anti-GOT1) at a 1:1,000 dilution (Abcam, ab171939), 620 1:1,000 dilution Anti-Rabbit LC3 A/B (CST, 12741), 1:1,000 dilution Anti-Rabbit NCOA4 (Bethyl 621 Laboratories, A302-272A), 1:1,000 dilution Anti-Rabbit Hif-2 (Abcam, ab131743), and loading 622 control vinculin at a 1:1,000 dilution (Cell Signaling, 13901), HSP-90 (Cell Signaling, 4877S),

24 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

623 Anti-Rabbit β-Actin (Cell Signaling, 4970L) or GAPDH (Cell Signaling, 2118). Anti-rabbit IgG, 624 HRP-linked (Cell Signaling Technology, 7074) secondary antibody was used at a 1:10,000 625 dilution. 626 627 Histology 628 Mice were sacrificed by CO2 asphyxiation followed by tissue harvesting and fixation overnight at 629 room temperature with Z-fix solution (Z-fix, Anatech LTD, #174). Tissues were the processed by 630 using a Leica ASP300S Tissue Processor, paraffin embedded, and cut into 5-μm sections. 631 Immunohistochemistry was performed on Discovery Ultra XT autostainer (Ventana Medical 632 Systems Inc.) and counterstained with hematoxylin. IHC slides were scanned on a Panoramic 633 SCANslide scanner (Perkin Elmer), and then annotation regions encompassing greater than 634 1mm of tissue were processed using Halo software (Indica Labs).The following antibodies were 635 used for IHC: GOT1 (AbCam, ab171939), Ki-67 (Cell Signaling, 9027), Cleaved Caspase-3 636 (Cell Signaling, 9664). 637 638 Metabolomics 639 Targeted metabolomics: Cells were plated at 500,000 cells per well in 6-well plates or ~1.5 640 million cells per 10 cm dish. At the endpoint, cells were lysed with dry-ice cold 80% methanol 641 and extracts were then centrifuged at 10,000 g for 10 min at 4°C and the supernatant was 642 stored at -80°C until further analyses. Protein concentration was determined by processing a 643 parallel well/dish for each sample and used to normalize metabolite fractions across samples. 644 Based on protein concentrations, aliquots of the supernatants were transferred to a fresh micro 645 centrifuge tube and lyophilized using a SpeedVac concentrator. Dried metabolite pellets were 646 re-suspended in 45 μL 50:50 methanol:water mixture for LC–MS analysis. Data was collected 647 using previously published parameters53,54. 648 649 The QqQ data were pre-processed with Agilent MassHunter Workstation Quantitative Analysis 650 Software (B0700). Additional analyses were post-processed for further quality control in the 651 programming language R. Each sample was normalized by the total intensity of all metabolites 652 to scale for loading. Finally, each metabolite abundance level in each sample was divided by the 653 median of all abundance levels across all samples for proper comparisons, statistical analyses, 654 and visualizations among metabolites. The statistical significance test was done by a two-tailed 655 t-test with a significance threshold level of 0.05. 656 657 Seahorse Mito Stress Test 658 MiaPaCa-2 cells were seeded at 2x104 cells/well in 80 μl/well of normal growth media (DMEM 659 with 25 mM Glucose and 2 mM Glutamine) in an Agilent XF96 V3 PS Cell Culture Microplate 660 (#101085-004). To achieve an even distribution of cells within wells, plates were incubated on 661 the bench top at room temperature for 1 hour before incubating at 37ºC, 5% CO2 overnight. To 662 hydrate the XF96 FluxPak (#102416-100), 200 μL/well of sterile water was added and the entire 663 cartridge was incubated at 37’C, no CO2 overnight. The following day, one hour prior to running 664 the assay, 60 μL/well of growth media was removed from the cell culture plate and cells were 665 washed twice with 200 μL/well of assay medium (XF DMEM Base Medium, pH 7.4 (#103575- 666 100) containing 25 mM Glucose (#103577-100) and 2 mM Glutamine (#103579-100)). After 667 washing, 160 μL/well of assay medium was added to the cell culture plate for a final volume of 668 180 μL/well. Cells were then incubated at 37’C, no CO2 until analysis. Also one hour prior to the 669 assay, water from the FluxPak hydration was exchanged for 200 μL/well of XF Calibrant 670 (#100840-000) and the cartridge was returned to 37’C, no CO2 until analysis. 671 Oligomycin (100 μM), FCCP (100 μM), and Rotenone/Antimycin (50 μM) from the XF Cell Mito 672 Stress Test Kit (#103015-100) were re-constituted in assay medium to make the indicated stock 673 concentrations. 20 μL of Oligomycin was loaded into Port A for each well of the FluxPak, 22 μL

25 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

674 of FCCP into Port B, and 25 μL of Rotenone/Antimycin into Port C. Port D was left empty. The 675 final FCCP concentration was optimized to achieve maximal respiration in each condition. 676 The Mito Stress Test was conducted on an XF96 Extracellular Flux Analyzer and OCR was 677 analyzed using Wave 2.6 software. Following the assay, OCR was normalized to cell number 678 with the CyQUANT NF Cell Proliferation Assay (C35006) from Thermo Fisher according to 679 manufacturer’s instructions. 680 681 Statistical analysis 682 Statistics were performed using GraphPad Prism 7 (Graph Pad Software Inc). Groups of 2 were 683 analyzed using the unpaired two-tailed Student’s t test and comparisons across more than 2 684 groups were conducted using one-way ANOVA Tukey post-hoc test. All error bars represent 685 mean with standard deviation, unless noted otherwise. A P value of less than 0.05 was 686 considered statistically significant. All group numbers and explanation of significant values are 687 presented within the figure legends. 688 689 Acknowledgements 690 The authors would like to thank members of the Olive, Shah, and Lyssiotis lab for critical 691 discussion and scientific feedback. D.M.K was supported by a Department of Education GAANN 692 fellowship through The University of Michigan Program in Chemical Biology. B.S.N was 693 supported by Institutional National Research Service Awards through NIDDK (T32-DK094775) 694 and NCI (T32-CA009676). M.P.M. was supported by NCI R01-CA-198074, R01-CA-151588, 695 and an ACS Research Scholar Grant. Y.M.S was supported by NIH grants (R01CA148828 and 696 R01DK095201). C.A.L. was supported by the Pancreatic Cancer Action Network/AACR (13-70- 697 25-LYSS), Damon Runyon Cancer Research Foundation (DFS-09-14), V Foundation for Cancer 698 Research (V2016-009), Sidney Kimmel Foundation for Cancer Research (SKF-16-005), the 699 AACR (17-20- 01-LYSS), and an ACS Research Scholar Grant (RSG-18-186-01). M.P.M., 700 Y.M.S., H.C.C., and C.A.L. were supported by the Cancer Center Support Grant (P30 701 CA046592). M.P.M., H.C.C., and C.A.L. were supported by U01-CA-224145. K.P.O. and C.A.L. 702 were supported by R01CA215607. Metabolomics studies at U-M were supported by DK09715, 703 the Charles Woodson Research Fund, and the UM Pediatric Brain Tumor Initiative. 704 705 Author Contributions 706 Conceptualization, D.M.K. and C.A.L.; Investigation, D.M.K, B.S.N., L.L., E.L.Y, A.M., G.T., 707 C.J.H., S.A.K., N.C., S.W., J.R., S.S., C.P., and T.G.; Resources, E.C., L.Z., M.A.B., P.S., 708 H.C.C., A.C.A., Z.C.N., J.M.A., M.P.dM., Y.M.S, and K.P.O.; Data Curation, D.M.K., P.S., A.C.A, 709 and L.Z.; Writing – Original Draft, D.M.K. and C.A.L.; Writing – Review & Editing, D.M.K., B.S.N, 710 E.L.Y., Z.C.N., Y.M.S., K.P.O., and C.A.L.; Formal analysis and Visualization, D.M.K; 711 Supervision and Funding Acquisition, C.A.L. 712 713 Declaration of Interests 714 C.A.L. is an inventor on patents pertaining to Kras regulated metabolic pathways, 715 redox control pathways in pancreatic cancer, and targeting GOT1 as a therapeutic approach.

26 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

716 Supplemental Information

27 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

717

28 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

718

29 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

719

30 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

720 721

31 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

722

32 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

723

33 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

724 References 725 1. Siegel, R. L., Miller, K. D. & Jemal, A. Cancer statistics, 2019. CA: A Cancer Journal for 726 Clinicians 69, 7–34 (2019). 727 2. Olive, K. P. et al. Inhibition of Hedgehog signaling enhances delivery of chemotherapy in a 728 mouse model of pancreatic cancer. Science 324, 1457–1461 (2009). 729 3. Feig, C. et al. The Pancreas Cancer Microenvironment. Clin Cancer Res 18, 4266–4276 730 (2012). 731 4. Zhang, Y., Crawford, H. C. & Pasca di Magliano, M. Epithelial-Stromal Interactions in 732 Pancreatic Cancer. Annual Review of Physiology 81, 211–233 (2019). 733 5. Koong, A. C. et al. Pancreatic tumors show high levels of hypoxia. International Journal of 734 Radiation Oncology*Biology*Physics 48, 919–922 (2000). 735 6. Perera, R. M. & Bardeesy, N. Pancreatic Cancer Metabolism: Breaking It Down to Build It 736 Back Up. Cancer Discov 5, 1247–1261 (2015). 737 7. Halbrook, C. J. & Lyssiotis, C. A. Employing Metabolism to Improve the Diagnosis and 738 Treatment of Pancreatic Cancer. Cancer Cell 31, 5–19 (2017). 739 8. Son, J. et al. Glutamine supports pancreatic cancer growth through a KRAS-regulated 740 metabolic pathway. Nature 496, 101–105 (2013). 741 9. Chakrabarti, G. et al. Targeting glutamine metabolism sensitizes pancreatic cancer to 742 PARP-driven metabolic catastrophe induced by ß-lapachone. Cancer & Metabolism 3, 12 743 (2015). 744 10. Anglin, J. et al. Discovery and optimization of aspartate aminotransferase 1 inhibitors to 745 target redox balance in pancreatic ductal adenocarcinoma. Bioorganic & Medicinal 746 Chemistry Letters 28, 2675–2678 (2018). 747 11. Holt, M. C. et al. Biochemical Characterization and Structure-Based Mutational Analysis 748 Provide Insight into the Binding and Mechanism of Action of Novel Aspartate 749 Aminotransferase Inhibitors. Biochemistry 57, 6604–6614 (2018). 750 12. Sun, W. et al. Aspulvinone O, a natural inhibitor of GOT1 suppresses pancreatic ductal 751 adenocarcinoma cells growth by interfering glutamine metabolism. Cell Communication and 752 Signaling 17, 111 (2019). 753 13. Yoshida, T. et al. A covalent small molecule inhibitor of glutamate-oxaloacetate 754 transaminase 1 impairs pancreatic cancer growth. Biochemical and Biophysical Research 755 Communications 522, 633–638 (2020). 756 14. O’Neil, N. J., Bailey, M. L. & Hieter, P. Synthetic lethality and cancer. Nature Reviews 757 Genetics 18, 613–623 (2017). 758 15. Zecchini, V. & Frezza, C. Metabolic synthetic lethality in cancer therapy. Biochimica et 759 Biophysica Acta (BBA) - Bioenergetics 1858, 723–731 (2017). 760 16. Mancias, J. D., Wang, X., Gygi, S. P., Harper, J. W. & Kimmelman, A. C. Quantitative 761 proteomics identifies NCOA4 as the cargo receptor mediating ferritinophagy. Nature 509, 762 105–109 (2014). 763 17. Li, C. et al. Identification of Pancreatic Cancer Stem Cells. Cancer Res 67, 1030–1037 764 (2007). 765 18. Nelson, B. S. et al. Tissue of origin dictates GOT1 dependence and confers synthetic 766 lethality to radiotherapy. Cancer & Metabolism 8, 1 (2020).

34 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

767 19. Dolma, S., Lessnick, S. L., Hahn, W. C. & Stockwell, B. R. Identification of genotype- 768 selective antitumor agents using synthetic lethal chemical screening in engineered human 769 tumor cells. Cancer Cell 3, 285–296 (2003). 770 20. Dixon, S. J. et al. Ferroptosis: An Iron-Dependent Form of Nonapoptotic Cell Death. Cell 771 149, 1060–1072 (2012). 772 21. Larraufie, M.-H. et al. Incorporation of metabolically stable ketones into a small molecule 773 probe to increase potency and water solubility. Bioorganic & Medicinal Chemistry Letters 25, 774 4787–4792 (2015). 775 22. Gorrini, C., Harris, I. S. & Mak, T. W. Modulation of oxidative stress as an anticancer 776 strategy. Nature Reviews Drug Discovery 12, 931–947 (2013). 777 23. Kamphorst, J. J. et al. Human pancreatic cancer tumors are nutrient poor and tumor cells 778 actively scavenge extracellular protein. Cancer Res. 75, 544–553 (2015). 779 24. Sullivan, M. R. et al. Quantification of microenvironmental metabolites in murine cancers 780 reveals determinants of tumor nutrient availability. eLife 8, e44235 (2019). 781 25. Meister, A. & Anderson, M. E. Glutathione. Annual Review of Biochemistry 52, 711–760 782 (1983). 783 26. Griffith, O. W. Mechanism of action, metabolism, and toxicity of buthionine sulfoximine and 784 its higher homologs, potent inhibitors of glutathione synthesis. J. Biol. Chem. 257, 13704– 785 13712 (1982). 786 27. Harris, I. S. et al. Deubiquitinases Maintain Protein Homeostasis and Survival of Cancer 787 Cells upon Glutathione Depletion. Cell Metab. 29, 1166-1181.e6 (2019). 788 28. Bailey, H. H. l-S,R-buthionine sulfoximine: historical development and clinical issues. 789 Chemico-Biological Interactions 111–112, 239–254 (1998). 790 29. Villablanca, J. G. et al. A Phase I New Approaches to Neuroblastoma Therapy Study of 791 Buthionine Sulfoximine and Melphalan With Autologous Stem Cells for Recurrent/Refractory 792 High-Risk Neuroblastoma. Pediatric Blood & Cancer 63, 1349–1356 (2016). 793 30. Yang, W. S. et al. Regulation of ferroptotic cancer cell death by GPX4. Cell 156, 317–331 794 (2014). 795 31. Stockwell, B. R. et al. Ferroptosis: A Regulated Cell Death Nexus Linking Metabolism, 796 Redox Biology, and Disease. Cell 171, 273–285 (2017). 797 32. Viswanathan, V. S. et al. Dependency of a therapy-resistant state of cancer cells on a lipid 798 peroxidase pathway. Nature 547, 453–457 (2017). 799 33. Skouta, R. et al. Ferrostatins inhibit oxidative lipid damage and cell death in diverse disease 800 models. J. Am. Chem. Soc. 136, 4551–4556 (2014). 801 34. Gaschler, M. M. et al. FINO2 initiates ferroptosis through GPX4 inactivation and iron 802 oxidation. Nat. Chem. Biol. 14, 507–515 (2018). 803 35. Doll, S. et al. FSP1 is a glutathione-independent ferroptosis suppressor. Nature 575, 693– 804 698 (2019). 805 36. Bersuker, K. et al. The CoQ oxidoreductase FSP1 acts parallel to GPX4 to inhibit 806 ferroptosis. Nature 575, 688–692 (2019). 807 37. Alvarez, S. W. et al. NFS1 undergoes positive selection in lung tumours and protects cells 808 from ferroptosis. Nature 551, 639–643 (2017).

35 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

809 38. Zou, Y. et al. A GPX4-dependent cancer cell state underlies the clear-cell morphology and 810 confers sensitivity to ferroptosis. Nature Communications 10, 1–13 (2019). 811 39. Thomas, F. et al. Calcein as a fluorescent probe for ferric iron. Application to iron nutrition in 812 plant cells. J. Biol. Chem. 274, 13375–13383 (1999). 813 40. Kimmelman, A. C. & White, E. Autophagy and Tumor Metabolism. Cell Metab. 25, 1037– 814 1043 (2017). 815 41. Yang, S. et al. Pancreatic cancers require autophagy for tumor growth. Genes Dev. 25, 816 717–729 (2011). 817 42. Sousa, C. M. et al. Pancreatic stellate cells support tumour metabolism through autophagic 818 alanine secretion. Nature 536, 479–483 (2016). 819 43. Zhou, X., Curbo, S., Li, F., Krishnan, S. & Karlsson, A. Inhibition of glutamate oxaloacetate 820 transaminase 1 in cancer cell lines results in altered metabolism with increased dependency 821 of glucose. BMC Cancer 18, 559 (2018). 822 44. Gao, M., Monian, P., Quadri, N., Ramasamy, R. & Jiang, X. Glutaminolysis and Transferrin 823 Regulate Ferroptosis. Mol. Cell 59, 298–308 (2015). 824 45. Gao, M. et al. Ferroptosis is an autophagic cell death process. Cell Res. 26, 1021–1032 825 (2016). 826 46. Hou, W. et al. Autophagy promotes ferroptosis by degradation of ferritin. Autophagy 12, 827 1425–1428 (2016). 828 47. Torti, S. V. & Torti, F. M. Iron and cancer: more ore to be mined. Nat. Rev. Cancer 13, 342– 829 355 (2013). 830 48. Gui, D. Y. et al. Environment dictates dependence on mitochondrial complex I for NAD+ and 831 aspartate production and determines cancer cell sensitivity to metformin. Cell Metab 24, 832 716–727 (2016). 833 49. Maddocks, O. D. K. et al. Modulating the therapeutic response of tumours to dietary serine 834 and glycine starvation. Nature 544, 372–376 (2017). 835 50. Mandal, P. K. et al. System x(c)- and thioredoxin reductase 1 cooperatively rescue 836 glutathione deficiency. J. Biol. Chem. 285, 22244–22253 (2010). 837 51. Gao, M. et al. Role of Mitochondria in Ferroptosis. Mol. Cell 73, 354-363.e3 (2019). 838 52. Magtanong, L. et al. Exogenous Monounsaturated Fatty Acids Promote a Ferroptosis- 839 Resistant Cell State. Cell Chem Biol 26, 420-432.e9 (2019). 840 53. Yuan, M. et al. Ex vivo and in vivo stable isotope labelling of central carbon metabolism and 841 related pathways with analysis by LC-MS/MS. Nat Protoc 14, 313-330 (2019). 842 54. Lee, H.J. et al. A large-scale analysis of targeted metabolomics data from heterogeneous 843 biological samples provides insights into metabolite dynamics. Metabolomics 15, 103 844 (2019).

36 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

845 Supplemental Figure Legends 846 847 Supplementary Figure 1, related to Figure 1. GOT1 dependence varies among 848 pancreatic cancers. A) Representative colony formation after 10-15 days (n=3) 849 corresponding to Figure 1D. B) Immunoblots corresponding to Figure 1D and SFigure 850 1A. Non-targeting, sh1, and sh3 stable cell lines were treated with dox for 5 days. VCL 851 was used as a loading control. C) Intracellular aspartate abundance in Pa-Tu-8902 852 shNT and sh1 stable cell lines after 5 days of dox-treatment accessed by liquid- 853 chromatography tandem mass spectrometry, normalized to mock (n=3). D) Relative 854 colony counts for NT (black/blue) stable cell lines and parental cell lines (grey/red) 855 corresponding to Figure 1D after 10-15 days (n=3). E) Relative cell number of 856 immortalized non-transformed human cell lines with shNT vectors corresponding to 857 Figure 1E. 1, 3, or 5-day time points were used, normalized to day 1 (n=3). F) Common 858 mutations associated with PDA from the Cancer Cell Line Encyclopedia. G) Expression 859 (mRNA) of malate-aspartate shuttle enzymes in GOT1 sensitive (n=7) and insensitive 860 (n=4) PDA cell lines from the Cancer Cell Line Encyclopedia. Sensitivity based on data 861 in Figure 1D. H) Growth of subcutaneous xenograft tumors containing non-targeting 862 (NT) vectors treated with dox (red) or vehicle (black) (n= 6), corresponding to Figure 1F. 863 I) Immunoblots for GOT1 corresponding to Figures 1J and SFigure 1H. VCL was used 864 as a loading control. J) Cell cycle distribution of Pa-Tu-8902 cells expressing shNT 865 constructs corresponding to Figure 1K. K) Relative cell number in iDox-shGOT1 or NT 866 cells corresponding to Figure 1H. 1, 3, and 5-day time points were used. Cell counts 867 were normalized to day 1 (n=3). Error bars represent mean ± SD. Non-significant P > 868 0.05 (n.s. or # as noted), P ≤ 0.05 (*), ≤ 0.01 (**), ≤ 0.001 (***), ≤ 0.0001 (****). See also 869 Figure 1. 870 871 Supplementary Figure 2, related to Figure 2. GOT1 knockdown sensitizes PDA to 872 xCT inhibition and cystine deprivation. A) Area under the curve (AUC) in cell viability 873 for each compound in Figure 2B for mock (black) or knockdown conditions (red) after 874 72 hours (n=3). B-D) Cell viability dose response curves for erastin after 24 hours in 875 iDox-shGOT1 expressing cell lines (B) depicted in Figure 2D. shNT expressing cell 876 lines were used in (C) and the AUC fold-change is presented, while imidazole ketone 877 erastin (IKE) was used in (D) (n=3). E) Relative MIA PaCa-2 iDox-shGOT1 cell numbers 878 after 5 days of dox treatment with the indicated conditions. 750nM of IKE was 879 administered on day 1 and each condition is normalized to day 1 (n=3). F) Relative 880 reduced glutathione levels (GSH) after 5 days of knockdown followed by 750nM of 881 erastin for 6 hours. GSH levels were first normalized to cell viability and then normalized 882 again to vehicle treatment (black and grey) (n=3). G) Cell viability of Mia PaCa-2 iDox- 883 shGOT1 after 5 days of dox culture then 750nM IKE co-cultured with the indicated 884 conditions (n=3). H) Relative Mia PaCa-2 iDox-shGOT1 cell numbers following 5 days 885 of GOT1 knockdown and the indicated media conditions (n=3). I) Cell viability of Pa-Tu- 886 8902 iDox-shGOT1 (red) and Mia PaCa-2 iDox-shGOT1 (blue) following 5 days of dox 887 pre-treatment and 24 hours of the indicated cystine concentrations (n=4). J) Total post- 888 treatment tumor burden (mock, n=5), (-Cys, 4), (dox, n=4), and (dox, -Cys, n=5). 889 K) Cysteine levels in tumors at endpoint (mock, n=5), (-Cys, 4), (dox, n=4), and (dox, - 890 Cys, n=6). Error bars represent mean ± SD (Figures S2a-h) or mean ± S.E.M (Figures

37 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

891 S2i-j). Two-tailed unpaired T-test or 1-way ANOVA: Non-significant P > 0.05 (n.s. or # as 892 noted), P ≤ 0.05 (*), ≤ 0.01 (**), ≤ 0.001 (***), ≤0.0001 (****). See also Figure 2. 893 894 Supplementary Figure 3, related to Figure 3. GOT1 knockdown sensitizes PDA to 895 GCLC inhibition. A-B) Dose-response curves for iDox-shGOT1 cells after 5 days of 896 GOT1 knockdown followed by BSO treatment for 24 (A) or 72 hours across 2 PDA cell 897 lines (B) (n=3). C) Area under the curve (AUC) normalized to +dox in matched sh non- 898 targeting (shNT). Cells were cultured with dox for 5 days prior to drug treatment.

899 Viability is normalized to a 0.1% DMSO vehicle control (n=3). D) Raw EC50 values 900 corresponding to Figure 3B and Figure S3B (n=3). E) Relative GSH levels normalized to 901 viability and vehicle after 6 hours of 40µM BSO treatment (n=3). F) Fold change of MIA 902 PaCa-2 iDox-shGOT1 cell numbers after 5 days of GOT1 knockdown and treatment 903 with the indicated conditions. 40µM of BSO was administered on day 1. Cell numbers 904 are normalized to day 1 for each condition (n=3). G) Relative viability of MIA PaCA-2 905 iDox-shGOT1 cells after 5 days of GOT1 knockdown and treatment with 40µM BSO or 906 co-treatment with 0.5mM N-acetyl cysteine (NAC) 0.5mM GSH-ethyl ester (GSH-EE, 907 n=3). H) Individual tumor volume measurements corresponding to Figure 3F (n=8). I) 908 Immunoblot analysis of GOT1 from tumors in Figure 3F (n=8). J) Relative abundances 909 of aspartate from whole tumor metabolite extracts from tumors in Figure 3E (n=8). Error 910 bars represent mean ± SD. Two-tailed unpaired T-test or 1-way ANOVA: Non-significant 911 P > 0.05 (n.s. or # as noted), P ≤ 0.05 (*), ≤ 0.01 (**), ≤ 0.001 (***), ≤0.0001 (****). See 912 also Figure 3. 913 914 Supplementary Figure 4, related to Figure 4. GOT1 suppression sensitizes PDA to 915 ferroptotic cell death. A) Percent cell viability dose-response curves from 5 additional 916 shGOT1 PDA cell lines upon 24 hours of RSL-3 treatment. GOT1 was knocked down 5 917 days prior to drug treatment. Viability is normalized to a 0.1% DMSO vehicle control

918 (n=3). B) Raw EC50 values for cells treated with RSL-3 for 24 hours. C) Area under the 919 curve (AUC) normalized to +dox in matched sh non-targeting (shNT) cell lines. Cells 920 were cultured with dox for 5 days prior to drug treatment. Viability is normalized to a 921 0.1% DMSO vehicle control (n=3). D) Fold change Mia PaCa-2 iDox-shGOT1 cell 922 numbers following 5 days of knockdown and treatment with the indicated conditions. 923 32nM of RSL-3 was administered on day 1. Cell numbers are normalized to day 1 for 924 each condition (n=3). E) Distribution of Pa-Tu-8902 iDox-shGOT1 cells positive for C11- 925 BODIPY corresponding to Figure 4E. Data are normalized to the -dox and vehicle- 926 treated condition (n=3). F) Relative lipid ROS in Pa-Tu-8902 iDox-shGOT1 treated with 927 32nM RSL-3 or 750nM Erastin -/+ 1μM Ferrostatin-1 (Fer-1) for 6 hours (n=3). 928 H) Single agent cell viability controls corresponding to Figures 5F and SFigure 5G. 929 Data are normalized to the -dox and vehicle treated control (n=3). I) Fold change Pa-Tu- 930 8902 iDox-shGOT1 cell numbers following 5 days of knockdown and treatment with the 931 indicated conditions (n=3). 32nM RSL-3, 750nM Erastin, or 40μM BSO -/+ 1μM 932 Ferrostatin-1 (Fer-1) were used. Cell numbers are normalized to day 1 for each

933 condition (n=3). J) Cell viability dose-response curves upon 24 hours of (–)–FINO2 934 treatment following 5 days of GOT1 knockdown (n=3). Viability is normalized to a 0.1% 935 DMSO vehicle control. Error bars represent mean ± SD. Two-tailed unpaired T-test or 1-

38 bioRxiv preprint doi: https://doi.org/10.1101/2020.02.28.970228; this version posted February 29, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

936 way ANOVA: Non-significant P > 0.05 (n.s. or # as noted), P ≤ 0.05 (*), ≤ 0.01 (**), ≤ 937 0.001 (***), ≤0.0001 (****). See also Figure 4. 938 939 Supplementary Figure 5, related to Figure 5. GOT1 knockdown promotes labile 940 iron release in response to metabolic stress. A) Model depicting known metabolic 941 regulators of ferroptosis. B) Immunoblot for HIF-2 in response to GOT1 knockdown. FG- 942 4592 is a positive control for HIF-2 expression. C-D) Gene expression (mRNA) of 943 HILPDA in Pa-Tu-8902 (C) and MIA PaCa-2 (D) following knockdown (n=4 in C, and 944 n=3 in D). E) Cell viability dose-response curves upon 72 hours of ferrous ammonium 945 citrate after 5 days of GOT1 knockdown (n=3). Viability is normalized to a 0.1% DMSO 946 vehicle control. F-G) Mean fluorescence intensity (MFI) (F) Calcein-AM in MIA PaCa-2 947 iDox-shGOT1 and relative labile iron (G) (n=4). H-I) mRNA expression of iron regulatory 948 genes in Pa-Tu-8902 (H) and MIA PaCa-2 (I) iDox-shGOT1 following 5 days of 949 knockdown (n=3). J) Representative images of basal autophagic flux in MT3-LC3 950 tandem fluorescence (GFP–RFP) reporter treated with GOT1 inhibitor PF-04859989. K) 951 AMP and ATP levels corresponding to Figure 5H measured by liquid-chromatography 952 tandem mass spectrometry, normalized to -dox (n=4). Error bars represent mean ± SD. 953 Two-tailed unpaired T-test or 1-way ANOVA: Non-significant P > 0.05 (n.s. or # as 954 noted), P ≤ 0.05 (*), ≤ 0.01 (**), ≤ 0.001 (***), ≤0.0001 (****). See also Figure 5. 955 956 Supplementary Figure 6. Representative C11-BODIPY Gating. Cells were first gated 957 by size using forward scatter area (FSA) vs. forward scatter height (FSH). Doublet 958 exclusion was then performed using side scatter area (SSA) vs. side scatter height 959 (SSH). Live cells were assessed selecting the Sytox Blue (DAPI) negative population in 960 both experiments. C-11 BODIPY (FITC) were measured in single and live cells for each 961 condition above. 962 963 Supplementary Figure 7. Representative Calcein-AM Gating. Cells were first gated 964 by size using side scatter area (SSC-A) vs. forward scatter area (FSC-A). Doublets 965 were excluded by comparing forward scatter width (FSC-W) vs. forward scatter height 966 (FSC-H). Live cells were assessed selecting the Sytox Blue (Pacific Blue) negative 967 population in both experiments. Calcein-AM (FITC) levels were measured in single and 968 live cells.

39